Risk stratification for arrhythmic death in an emergency department cohort: a new method of nonlinear PD2i analysis of the ECG

Figures & data

Table 1 Clinical characteristics of study cohort

Clinical characteristic	Enrolled patients1 N = 918	Lost to follow-up2 N = 42	No ECG or consent3 N = 57
Average age	63	51	66
Average LVEF	50	52	47
% Male	52*	75*	58
% Diabetic	23	31	25
% Hypertension	57	53	55
% CHF	10	7	11
% Prior MI	25	21	28
% Current AMI	21	14	26

1 Caucasian 73%, Afro-American 24%, [Hispanic (all nonAfro) 7%], Asian 3%.

2 Lost to follow-up (4.6%); 36 PD2i-Neg; 3 PD2i-Pos, 3 PD2i-Rej.

3 No ECG recorded or patient withdrew consent (6.2%).

* Different from enrolled, p ≤ 0.001 (Chi-square); all others are nonsignificant.

Abbreviations: AMI, acute MI; CHF, congestive heart failure; MI, myocardial infarction; LVEF, left ventricular ejection fraction.

Table 2 Contingency table for primary results

ER study
6 Tertiary-care hospitals
918 Total enrolled, 819 completed study
	AD	no AD
PD2i Pos	26 TP	103 FP
PD2i Neg	1 FN	595 TN
Sensitivity = 26/(26 + 1) = 96%*
Specificity = 595/(595 + 103) = 85%*
Negative predictive value = 595/(595 +1) = 99%*
Positive predictive value = 26/(26 + 103) = 20%
Relative risk = 26/1 ×(595 + 1)/(26 + 103) = 24.2*

* p < 0.01, Fisher’s Exact Test for contingency tables.

Abbreviations: AD, arrhythmic death; TP, true positive; FP, false positive; FN, false negative; TN, true negative.

Table 3 Clinical characteristics and outcome distributions

Clinical characteristic	All patients1	PD2i positive		PD2i negative
		Alive2	Died3	Alive4	Died5
% PD2i Test**	100	13	3	73	0.1
Average age*	63	62	72	61	73
Average LVEF	50	52	34	50	None
% Male	52	6	2	44	0
% Diabetic	23	4	1	18	0
% Hypertension	57	9	2	46	0
% CHF	10	3	1	6	0
% Prior MI	25	3	1	21	0
% Current AMI**	21	1	2	18	0.1

1 819 with complete data; 11% had PD2i Test rejection (2 ADs).

2 103 False positives

3 26 True positives

4 595 True negatives

5 1 False negative

* p ≤ 0.01,

** p ≤ 0.001 (Cox Proportional Hazards Model)

Abbreviations: AMI, acute MI; CHF, congestive heart failure; MI, myocardial infarction; LVEF, left ventricular ejection fraction.

Table 4 Test for homogeneity of enrollment groups

Breslow-Day Test for homogeneity of the odds ratios using three scenariosa
Chi-square	S 0.9261
	C 2.5088
	B 1.4552
DF	S 1
	C 1
	B 1
Pr > Chi-square	S 0.3359b
	C 0.1132b
	B 0.2277b

a S = Standard; C = Conservative; B = Best.

b not statistically significant.

Figure 1 Receiver-operator curve (ROC) for all data with a positive or negative PD2i Test. The ROC determines the criterion cut-point for the PD2i Test that maximizes the sensitivity and specificity. At a cut-point of PD2i ≤ 1.4, the maximum sensitivity and specificity are found with the area under the ROC being high (0.92).

Figure 2 RR intervals and associated PD2is from two types of patients. Upper panel shows RR data and corresponding PD2i results from a patient who experienced documented arrhythmic death (AD) within 24 hours; note the two patterns of either a sustained low-dimension (line) or the systematic low-dimensional excursion (arrow). The lower panel shows data and results from a normal patient discharged after diagnosis of gastro-esophageal reflux disorder (CONTROL).

Figure 3 All-cause death vs time of occurrence after the brief ECG recording. Large squares indicate cardiac arrhythmic death (AD), as adjudicated by the Events Committee. All non-AD deaths are indicated by small unfilled squares, and include deaths due to cancer, sepsis, stroke, etc. AD patients admitted to low-level telemetry observation at admission are indicated by dots.

Figure 4 Composite RR and PD2i results for the first 18 arrhythmic death (AD) cases with all clinical data and no a priori exclusions for noise contamination. NCA = cases in which a single bit (noise-bit) was removed from each data point thus forcing the background noise level below ± 5 integers. In the acute myocardial infarction (AMI) controls all PD2i >3.0 are pushed down to the 3.0 line for display.